http://rfid.cs.washington.edu/ the rfid ecosystem project longitudinal study of a building-scale...

http://rfid.cs.washington.edu/The RFID Ecosystem Project

Longitudinal Study of a Building-ScaleRFID Ecosystem

Evan Welbournewith

Karl Koscher, Emad Soroush, Magdalena Balazinska, Gaetano Borriello

University of Washington, CSE

MobiSys 2009June 22, 2009 - Kraków, Poland

http://rfid.cs.washington.edu/Focus: Pervasive RFID Systems

RFID tags on people and objects Higher-level events are inferred

http://rfid.cs.washington.edu/

tag Antenna Time

RFID Trace:

Bob, notes C 1

Bob, notes B 4

Bob, notes A 6

… … …

“Working in Office”

http://rfid.cs.washington.edu/Focus: Pervasive RFID Systems

RFID tags on people and objects Higher-level events are inferred High-performance passive tags

http://rfid.cs.washington.edu/

[ http://www.pcts.com]

Detect CareMilestones

[ http://www.aeroscout.com]

Track Hospital’s Equipment, Staff

[ http://www.pcts.com]

Active Tags Passive Tags Battery-powered $10 - $100 (US) Reliable location stream

No batteries < $1 (US) Less reliable location

VS.

EPC Gen 2 tags

What does the data from a pervasive system look like? How well do the tags perform? Can performance be improved? Will users adopt passive RFID tags? Do users accept applications built on passive RFID? Past studies are limited to lab-like settings…

http://rfid.cs.washington.edu/

4-week study of a building-scale RFID deployment: EPC Gen 2 RFID 67 participants 300+ tags Location apps

Summary: 1.5M tag reads 38,000 antenna visits 9 lost or broken tags 9,000 application operations by participants 0 reported privacy breaches

Longitudinal Study

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

CSE Building (8,000 m2) 47 Readers, 160 Antennas Installed in 3 configurations

The RFID Ecosystem

First Floor Entrance(not shown on map)

http://rfid.cs.washington.edu/

Three tag designs

324 tags on 19 types of objects:

RFID Tags

Personal Badges Bags Clothing Keys Wallets, Purses

Books Paper, binders iPods, Laptops, Phones Food / Water Containers And more…

http://rfid.cs.washington.edu/

Tag Read Events (TREs) and STAYs TRE: (tag, antenna, time) STAY: (tag, antenna, start time, stop time)

Example:

RFID Data Streams

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/Data Rates and System Load

Unlike supply chain RFID applications…

Amount of data generated is very manageable

Total raw data: 53 MB Compresses to: 1 MB

Peak Rate: 8400 TRE / hour

Scale to 1M tags: ~ 100 MB / day

Explanation:

Fewer tags than in the supply chain

No antennas inside offices

Many tagged objects never move

Supply chain numbers may not be compressed

http://rfid.cs.washington.edu/Data Rates and System Load

Like similar studies in wireless mobility…

Load mirrors patterns of building occupancy

US Veteran’s Holiday US Thanksgiving Holiday

Major Undergrad Project Due

Implications: Makes sense to allocate more resources to hot spots

Predictable off-peak times for batch processing

“Hot spot” antennas: Outside research lab Outside student store

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

1358 web-surveys collected ground truth location data

We compute Detection-Rate:

Probability that a tag is detected by a nearby antenna

Tag Performance

http://rfid.cs.washington.edu/

Key Factors:

Tag Performance

Compared to performance in controlled laboratory studies…

Performance was significantly worse

1) Tag design 2) Object type

ex: wallet, purse, keys

More RF-absorbent

Tags held tight against body

ex: bag, hat, helmet, jacket

Less RF-absorbent

Tags more separated from body

High variance:Differences in tag mounting

(Also slight differences in material composition)

Implications: Must cope with high uncertainty in raw data Performance engineering opportunities:

3) Mounting

Larger antennas work betterAll lower than detection rate in lab studiesOther factors: simultaneous mobile tags, RF interference,… Best design Better mounting Redundancy*

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

Tackle inherent uncertainty of event detection as well

Approach: Probabilistic model of location w/particle filter Process with probabilistic event detection engine

Evaluation: “Entered-Room” using survey results

Despite high uncertainty: 60% correct room, 80% correct vicinity But doesn’t work when detection rate < 0.5 Probabilistic data is computationally expensive

Probabilistic Inference

To further improve performance on top of uncertain RFID data…

Probabilities help, must be applied with care

?

?

?

?

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

Participants mounted & managed their own tags Only basic guidelines and reminders for 3 weeks Experts remounted poorly performing tags in week 4

Exit survey for 67 participants also showed:

41 often forgot to carry tags

8 found tags socially awkward to carry/wear

19 said tags were inconvenient or uncomfortable

9 tags reported lost or broken

End-User Tag Management

To encourage optimal tag performance…

Tag use must be supported and incentivized

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

Deployment overview

Data rates and system loadWhat does the data from a pervasive system look like?

Tag PerformanceHow well do passive tags perform in practice?

Probabilistic InferenceCan performance be improved in software?

End-user tag managementWill users adopt passive tags?

ApplicationsDo users accept applications based on passive RFID?

Outline

http://rfid.cs.washington.edu/

Standard location-aware applications Built using Cascadia system (MobiSys 2008)

Event detection over deterministic raw data

Applications

http://rfid.cs.washington.edu/

Tool for deleting data was used “only as a test” Relatively few operations to manage privacy Many said: “Not concerned about tracking in CSE”

“More concerned about employer or gov’t”

Only 7 participants removed tags for privacy reasons 12 reported behavior change (e.g., arriving earlier) 0 reported privacy breaches

[See IEEE Pervasive 07, Internet Computing 09 for more on privacy]

Applications – Privacy

Contrary to expectation…

Very few participants concerned w/personal privacy

http://rfid.cs.washington.edu/

Exit-survey for 67 participants showed: Most found apps novel, fun; 15 found them truly useful Barriers: log-in (18), lack of friends (15), poor tags (5)

Requested features: Push-based interfaces (Desktop Feed gadget) More complex events, historical data (“Digital Diary”)

Applications - Adoption

Applications were “novel, fun, useful” depending on…

Ease of use, participating friends, tag performance

Future Work: Push information to users rather than pull

Support more complex events, historical queries

http://rfid.cs.washington.edu/

Through longitudinal study in a pervasive deployment:

Data rates and system load are quite manageable

Tag performance worse than in lab but there’s hope

Probabilistic inference helps but must be applied w/care

End-users need substantial support in mounting tags, applications are a good incentive

Users accept applications (when they work), but more sophisticated functionality is desired

Conclusion:

Pervasive computing with passive RFID is feasiblebut extensive optimizations are required

Summary

http://rfid.cs.washington.edu/

Thank you!

See our website for more information:

http://rfid.cs.washington.edu/

http://rfid.cs.washington.edu/publications.html

http://rfid.cs.washington.edu/Backup Slides

Backup Slides…

http://rfid.cs.washington.edu/

3 types of tags:

Lab benchmarks showing read-rate:

EPC Gen 2 RFID Equipment

FleXwing Excalibur PVC Card

http://rfid.cs.washington.edu/

Email sent to Faculty, Staff, grads, undergrads $30 for participation

+ $10 for completing >= 25% of surveys + $20 for completing >= 50% of surveys

Recruiting Participants

Recruited ParticipantsFaculty Participants 2

Staff Participants 2

Grad Participants 30

Undergrad Participants 33

MALE 46

FEMALE 21

http://rfid.cs.washington.edu/Hourly Data Rates

Total: 1.5M TREs 38K STAYs

Max Rate: 8408 TRE/hr 601 STAY/hr

Min Rate: 0 TRE/hr 0 STAY/hr

http://rfid.cs.washington.edu/People and Objects

Participants: Almost no data or a lot of dataSome forgot tags often; Some not in bldg

Objects: Same trendMobility of object; Material of object

http://rfid.cs.washington.edu/Load Distribution

Some antennas are “hot spots”Similar to wifi mobility studies

http://rfid.cs.washington.edu/Survey Responses

2226 Surveys Sent, 1358 Received 18 seconds to complete (avg)

http://rfid.cs.washington.edu/

Usage varies by participant 0.37 correlation: data generated and app usage

Application Usage

http://rfid.cs.washington.edu/

Use Historical Data View Trends

More Recent Applications

http://rfid.cs.washington.edu/

Ambient AwarenessFriend and Object FindersTime Use Analysis ToolsContext-Aware Social NetworkingRFID-based Reminders

Supported by Cascadia System

Welbourne, E. et al., MobiSys 2008

http://rfid.cs.washington.edu/

Privacy Tools: Data review tool Access control tool

Privacy Tools

http://rfid.cs.washington.edu/Issue: Basic Insecurity of RFID

Case Study: WA State Enhanced Driver’s License

DHS claims RFID “removes risk of cloning” Can be cloned easily in less than a second w/cheap device

Can be read more than 75 ft away

Sleeve doesn’t always work, worse when crumpled

# EDL Reads, Week of Apr 27th

Case study credit: Karl Koscher, Ari Juels, Tadayoshi Kohno, Vjekoslav Brajkovic